Crypto Hacks Drop to $26.5M in February Amid Security Gains

TheNewsCrypto发布于2026-03-02更新于2026-03-02

文章摘要

Crypto hacks and scams plummeted to a multi-year low in February, totaling approximately $26.5 million across 15 incidents. This represents a significant 69.2% decrease from January's $86 million in losses. The two largest attacks accounted for the majority: a $10 million price manipulation hack on YieldBlox’s lending pool and an $8.9 million private key exploit on IoTeX. Security firm PeckShield attributed the sharp decline to a lack of "mega-hacks" and a shift in focus due to market volatility. Analysts suggest capital is becoming more selective, rewarding protocols with stronger security, and that improved risk controls, audits, and AI-powered tools are contributing to a broader trend of enhanced crypto security.

Last month recorded the lowest level of crypto hacks and scams since March 2025, having $26.5 million stolen since February, as reported by the blockchain security company PeckShield.

Around 15 cases were reported in the month, in which two accounted for the majority of the losses, with the biggest accounting for the $10 million theft from YieldBlox’s DAO-managed lending pool through a price manipulation hack on February 21, as reported by PeckShield at X on March 1.

The second-biggest hack targeted the decentralised identity protocol IoTeX, which lost around $8.9 million to a private key exploit on Feb 21. In total, February indicates a 69.2% month-on-month fall from January, which listed more than $86 million in losses.

What Did the Spokesperson Mention?

A spokesperson from PeckShield mentioned that mega-hacks, like the $1.5 billion Bybit hack in February last year, did not inflate last month’s statistics, and market volatility resulted in a significant cooling period in exploit activity.

The spokesperson further mentioned that a sharp market correction in early February, having Bitcoin slip below $70,000, moved the focus of the industry toward institutional deleveraging and math-based sell-offs.

At the time of high-volatility periods, the tactical aim mostly shifts away from protocol exploits toward navigating market liquidity. Dominick John, a Kronos Research analyst, mentioned that the shift could also show tighter risk controls, stronger counterparty standards and amplified real-time monitoring over major venues.

He further went on, mentioning that Capital is shifting to become more selective, rewarding protocols with mature security frameworks. John mentioned losses could carry on to slip by the year as audits, monitoring, and institutional risk frameworks mature.

AI might also intensify the shift, backing automated code reviews, anomaly detection, and pre-deployment attack simulations to catch vulnerabilities earlier in the lifecycle. Crypto security is intensifying, and protocols are doubling down on audits, formal verification, and real-time monitoring.

Highlighted Crypto News Today:

Will AAVE Reset Monthly Momentum or Lose Steam and Remain in the Red?

TagsBybitCrypto AttackHack

相关问答

QWhat was the total amount stolen in crypto hacks and scams in February, as reported by PeckShield?

A$26.5 million.

QWhich two major hacks accounted for the majority of the losses in February, and what were the amounts?

AThe $10 million theft from YieldBlox’s DAO-managed lending pool and the $8.9 million hack targeting the IoTeX protocol.

QWhat was the month-on-month percentage decrease in crypto losses from January to February?

AA 69.2% decrease.

QAccording to the spokesperson, what factor contributed to a significant cooling period in exploit activity?

AMarket volatility.

QHow might AI contribute to improving crypto security, as mentioned in the article?

ABy backing automated code reviews, anomaly detection, and pre-deployment attack simulations to catch vulnerabilities earlier.

你可能也喜欢

生成模型也能端到端训练了?核心竟是一个for循环

长期以来,深度学习在图像分类、检测等任务上凭借端到端训练取得了巨大成功,但生成模型领域却始终是个例外。当前主流的自回归、扩散模型等训练时只学习预测“一小步”,推理时则需要反复展开数百上千步,这种训练与推理的不一致导致了“暴露偏差”等问题,使得生成模型一直未能实现真正的端到端训练。 最近,一篇来自UIUC与哈佛大学的论文提出了“探索式建模(Explorative Modeling,XM)”新范式,其核心是一个简单到近乎朴素的for循环:在每个训练步骤中,模型生成K个候选样本,然后仅选取其中最接近真实数据的一个样本来计算损失和回传梯度。这种做法巧妙地绕开了传统生成模型中因使用重构损失(如平方误差)而导致的“模态模糊”问题——即模型倾向于输出多个可能答案的平均值,而这个平均值往往不属于任何真实的数据模式。 XM通过这种方式,在不拆分生成过程的前提下,直接提升了模型的“生成表达力”,即模型捕捉数据分布中多个独立模式的能力。论文将“探索”验证为继模型参数量、数据量之后的第三根可扩展轴。实验表明,随着模型规模、数据量和计算量的增大,探索带来的性能收益越发显著,在图像、视频、语言等多种任务上均能带来效率与质量的提升。 更重要的是,当探索程度足够时,XM可以实现真正的端到端生成。在机器人控制等任务中,仅需一次网络前向传播的XM策略,其性能可媲美甚至超过需要上百次前向的扩散模型策略,实现了推理效率的飞跃。 尽管“最佳K样本选择”的思想并非全新,但该研究的贡献在于清晰地揭示了这一简单机制如何直接解决生成模型的核心瓶颈,并为实现高效、端到端的生成模型开辟了新的可能性。随着模型规模的持续扩大,探索这一新维度可能将发挥越来越关键的作用。

marsbit8分钟前

生成模型也能端到端训练了?核心竟是一个for循环

marsbit8分钟前

年薪百万抢电工,Meta急到自己办技校

AI竞赛正面临新的瓶颈:工地。美国正面临严重的电工、建筑工等技术工人短缺,这已成为微软、Meta、OpenAI等公司快速建设超大规模AI数据中心(如OpenAI耗资160亿美元的“星际之门”项目)的“头号障碍”。 尽管AI公司愿支付高薪(如电工年薪可达24-28万美元),远超传统行业,但熟练技工仍供不应求。麦肯锡预测,美国在2023-2030年间需额外培养13万名电工和24万名建筑工,而劳工统计局预计每年仍有8万个电工岗位空缺。这种短缺导致项目延迟,每月可能造成数百万美元的收入损失。 AI数据中心建设复杂,需应对巨大功耗(一座设施耗电堪比数十万户家庭)、复杂的配电系统以及高密度散热(需液冷技术)等挑战,因此亟需大量技术娴熟的工人。 为此,科技巨头开始亲自下场培养人才。例如,Meta投入1.15亿美元建立建筑工人培训学校,提供免费培训及生活补贴,以快速输送工人上岗。OpenAI则与建筑工会合作,提前锁定熟练劳动力。同时,企业也将招聘目光投向高中生,鼓励年轻人投身技工行业。这些举措已见成效,Z世代对技工职业的兴趣显著上升。 然而,更深层的挑战在于电力。AI数据中心用电量正以惊人速度增长,已推高部分地区的电价。此外,数据中心建设是项目制的,建设期需要成千上万的工人,但建成后仅需少量常驻人员。这意味着未来可能面临熟练工人短期过剩、并流向其他行业压低薪资的风险。如何实现劳动力与电力资源的长期平衡,仍是悬而未决的问题。

marsbit1小时前

年薪百万抢电工,Meta急到自己办技校

marsbit1小时前

交易

现货
活动图片